Semiconductor Industry Splits On 450mm Wafers - HotHardware
Semiconductor Industry Splits On 450mm Wafers

Semiconductor Industry Splits On 450mm Wafers

Back in May, Intel partnered with TSMC and Samsung to announce their intention of developing and deploying 450mm wafer starts by 2012. The shift from 300mm (12" diameter wafers) to 450mm was touted as necessary for all the usual reasons—think cheaper, faster chips—and we've not heard much in the way of pushback until now. At the Semicon West conference this week, a number of companies voiced their belief that a shift to 450mm wafers was premature, unproven, and would impose a tremendous financial burden.

There are two basic ways a semiconductor manufacturer can attempt to increase production capacity and revenue. The first (and generally simpler) approach is to shrink the company's existing manufacturing process. All else equal, a chip built on a 45nm process is smaller and typically draws less power than the exact same processor produced at 65nm or 90nm. As the manufacturing process shrinks, the number of CPU cores-per-wafer increases and the manufacturing cost of each chip goes down (again, we're assuming equal yields at both processes).

The other way to boost production capacity is to increase wafer size. Doing so deals with the capacity issue more directly, but also carries a prohibitively high price tag. In order to accommodate the larger wafers, a company must upgrade the entire assembly line. There's no such thing as a gradual or transitionary period—either the entire line accepts the larger wafer size, or it does not. Intel and its partners claim that by backing 450mm technology they're innovating through current financially uncertain times, while the company's competitors see things rather differently.

Global Foundaries Fab 1 - Dresden, Germany

"The rush to 450-mm suggests a lack of ideas for improving fab productivity," said GlobalFoundries VP of manufacturing systems and technology Thomas Sonderman. "At GlobalFoundries, we see a tremendous amount of headroom left in the 300-mm process." With all due respect to Mr. Sonderman, that's precisely the response we'd expect from any company that can't reasonably expect to shoulder the cost of a 450mm conversion in the near future. Both the Fab 38 conversion GF completed earlier this year and the New York State fab it intends to build are based around 300mm wafers. The current state of the global economy combined with a distinct lack of interest from a number of semiconductor firms could combine to push 450mm wafer adoption out well past the current 2012 goal. Intel may be willing to shoulder the burden for various reasons that we'll touch on in another update, but companies like UMC, Chartered, and Globalfoundries could wait another six-to-eight years before deciding to take the plunge themselves.
+ -

>> In order to accomodate the larger wafers, a company must upgrade the entire assembly line. There's no such thing as a gradual or transitionary period—either the entire line accepts the larger wafer size, or it does not. <<

What if you started the upgrades from the end of the line, in a manner that supports variable sizes? You could gradually upgrade everything, then once the whole line is done you throw a few 450's in the mix to make sure it all works. If there's a problem with any individual components, pull out the 450's and continue running 300's until you fix that one component.

I may be thinking too abstract; there might be some reason specific components can't accommodate variable sizing.

+ -

IMO there are several other non-technical topics also related to 450mm conversion.  The first topic is that we are still dealing with the industry oversupply and consolidation resulting from the transition to 300mm.  Unless total semiconductor wafer consumption increases at a faster rate, 450mm drives further industry consolidation.

The second topic is that some suggest lithographic and basic cell structure challenges increase substantially as we approach 20nm processes.  Does the industry want to resolve those issues before, during, or after the 450 mm conversion?  My guess is that manufacturing challenges related to logic may be easier to resolve compared to memory, analog, and SOC companies.  If that is the case, the industry will not be uniformly motivated to convert to 450mm.

The third topic is who will pay for the research necessary to create 450mm fabs--particularly if the benefits are initially enjoyed by a declining percentage of large companies?

+ -


I can't directly answer your question other than to observe that no one has ever taken this approach that I'm aware of. Since the industry has moved from one wafer size to the next multiple times in the past 20 years, it seems reasonable that there's compelling reason not to do so.


Take your idea and extend it a little, and you'll see why Intel really, really wants a future where Atom populates a huge number of devices. The current-generation Atom is about 25mm sq and built on a 300mm wafer. Imagine how many 28nm Atom's Intel could fit on a 450mm wafer, plus all the added cost of moving to said wafer size, and it's obvious that Intel is planning on moving a *lot* of these little guys 4-6 years down the road.

Right now, Intel can put an Atom in a netbook. In the future, Intel wants to put an Atom in your cell phone. If Intel can put an Atom in your cell phone, how long before it starts looking around to put an Atom in virtually everything else?

Login or Register to Comment
Post a Comment
Username:   Password: